FY 2026

Assessing the observational uncertainties of dissolved oxygen climatology and seasonal cycle through a coordinated intercomparison project

Ito, T., H.E. Garcia, Z. Wang, L. Cheng, J. Du, C.J. Roach, Y. Zhou, J.D. Sharp, S.K. Lauvset, S. Minobe, S. Bushinsky, B. Lu, and G. Giacomo Navarra

Global Biogeochem. Cycles, 39, e2025GB008751, doi: 10.1029/2025GB008751, View open access article at AGU/Wiley (external link) (2025)

Uncertainties in global ocean oxygen inventories are assessed by a coordinated intercomparison of dissolved oxygen inventories derived from two observational data sets with distinct quality control (QC) protocols and five different statistical interpolation methods. We investigate key sources of uncertainty including mapping interpolation schemes and data QC methods, which contribute more significantly than measurement or sampling errors. Local differences in mapped oxygen content can reach up to 10 μmol/kg (about 4% of the surface climatological mean), especially in the regions of high variability and poor sampling such as the eastern tropical Pacific and the coastal Antarctica. Globally integrated differences, however, are small (≤0.17% above 2,000 m depth). Mapping methods are likely the largest contributor of the uncertainty for the annual mean, but both mapping and QC methods are important for the seasonal cycle. These results are limited by only including two sets of QC methods and only statistical interpolation techniques. Future incorporation of machine learning-based methods and time-dependent oxygen maps will be critical for tracking deoxygenation trends and for providing observational constraints to validate Earth System Models.

Plain Language Summary. This study looked at how much uncertainty there is in our observation-based estimates of the oxygen stored in the ocean. We compared results from two different ways of checking data quality, and five different methods for filling in gaps where measurements are missing. The biggest differences came from the ways used to fill in those gaps more than from errors in the actual oxygen measurements. In some places, especially areas like the eastern tropical Pacific and coastal Antarctica where the ocean changes more vigorously and fewer measurements exist, the differences in oxygen estimates could be large. However, when averaging those differences over the entire ocean waters, the differences became very small. The way the data was filled in still had the most impact on yearly averages. The study had some limitations, such as not including newer methods like machine learning. Using those new methods in the future can broaden the comparison.